Eli Lilly's $2.75B AI Drug Deal: Pharma's Billion-Dollar Bet on Insilico

Your weekly signal boost from 190,000+ articles, served with a DJ's ear for what actually matters.

So, What Actually Happened?

We scanned 190,000 articles this week so you don't have to. And the signal that cut through the noise? AI stopped being a slide deck topic and started writing real checks. Eli Lilly signed a deal worth up to $2.75 billion with Insilico Medicine for AI-powered drug discovery, the largest pharma-AI partnership ever announced. Meanwhile, a startup called Starcloud raised $170 million to build data centers in orbit and immediately hit unicorn status. Back on earth, Palantir quietly renewed its AI partnership with automaker Stellantis for five more years, and Australia's largest bank is co-designing a next-generation security platform that ingests 30 terabytes of threat data per day.

The Bottom Line: AI verticalization has arrived. The companies writing billion-dollar checks are not buying potential. They are buying proven, industry-specific AI that already delivers results.

The Tracks That Matter

1. Eli Lilly Just Bet $2.75 Billion That AI Can Find Drugs Faster Than Humans. The Deal Size Tells You They Already Know It Works.

Eli Lilly has struck a deal worth up to $2.75 billion with Hong Kong-based Insilico Medicine to use AI across the entire drug discovery process, from identifying biological targets to designing the compounds themselves. This is not a research grant or a pilot program. This is a $35 billion pharmaceutical giant telling the market that AI drug discovery has crossed the threshold from experimental to operational.

The numbers behind Insilico explain the conviction. Insilico has developed more than 28 drugs using generative AI, with nearly half already in clinical-stage trials. The company went public in Hong Kong in December and its stock is up more than 50% year-to-date. Andrew Adams, Lilly's group vice president of Molecule Discovery, called Insilico's AI platform ”a powerful complement” to Lilly's own clinical development work. Investopedia reports this is an expansion of an existing relationship, not a cold start. Lilly is doubling down because the first phase delivered.

What makes this deal structurally significant is the China dimension. Eli Lilly is simultaneously investing $3 billion in China over the next decade, a market that currently accounts for less than 3% of its revenue. CEO David Ricks attended a high-level forum in Beijing, underlining the company's growing focus on China-linked opportunities. While governments debate AI export controls, Lilly is building research bridges. BioSpace reports that the partnership gives Lilly direct access to one of the most advanced AI drug discovery platforms currently operating, with a portfolio already deep into clinical testing.

Here's what works: If your organization is evaluating AI partnerships, study the Lilly playbook. They started small, validated results, then scaled to $2.75 billion. That is the pattern. Before committing to any AI vendor, ask: ”What has this platform already delivered in production, not in demos?” If the answer involves pilot programs and proofs of concept, you are still in the tire-kicking phase. Lilly moved past that.

2. A Startup Just Raised $170 Million to Build AI Data Centers in Orbit. It Reached Unicorn Status on Day One.

Starcloud secured $170 million in Series A funding to launch orbital AI data centers in low Earth orbit, reaching a $1.1 billion valuation and unicorn status with a single raise. The company's thesis is provocative but logical: terrestrial data centers face binding constraints on energy, cooling, and physical space. Orbit has none of those problems.

This sounds like science fiction until you look at what is driving it. AI workloads are consuming electricity at rates that strain national grids. Companies are competing for real estate near power substations. Cooling costs are rising. The bottleneck is no longer the chip or the algorithm. It is the physical environment in which the chip operates. Starcloud's bet is that solving the environment problem in orbit is cheaper than solving it on the ground, especially at the scale AI infrastructure is heading toward.

The valuation is the tell. A startup does not reach unicorn status on a Series A without investors seeing a path to revenue that terrestrial alternatives cannot match. This is the same pattern we tracked with infrastructure plays like Kandou AI's $225 million raise for interconnect technology: capital is flowing to the layer that sits beneath the models. The models get the headlines. The infrastructure gets the checks.

Here's what works: You do not need to start planning for orbital computing tomorrow. But you do need to track where data center constraints are creating new markets. If energy, cooling, or physical space is limiting your AI deployment roadmap, monitor the companies solving those constraints. The ones that succeed will set the terms for the next decade of AI infrastructure pricing.

3. Australia's Largest Bank Is Co-Designing a Security Platform That Ingests 30 Terabytes of Threat Data Per Day. The Word 'Agentic' Is Doing the Heavy Lifting.

National Australia Bank has become one of five design partners for Lakewatch, a new agentic security platform where custom security agents can be built and run to handle incident detection and response tasks. NAB currently ingests more than 30 terabytes of security data each day. That volume alone makes traditional rule-based security monitoring impractical. The platform integrates data from 15 security vendors into a unified layer.

The word ”agentic” matters here. This is not a dashboard refresh. An agentic SIEM means autonomous agents can detect threats, correlate signals across data sources, and execute response workflows without waiting for a human analyst to notice an alert. When your bank processes 30TB of daily security data, the question is not whether AI should handle the first line of defense. The question is whether human analysts can handle it at all.

NAB's chief security officer Sandro Bucchianeri put it directly: ”The ability to work in collaboration to shape a product to meet the needs of cyber defenders into the future excites us.” That language, ”shape a product,” tells you NAB is not buying a finished tool. They are building one alongside the platform maker. The design partner model gives NAB influence over what the product becomes, while giving the platform real-world validation at enterprise scale.

Here's what works: If your security team is still evaluating SIEM replacements, understand that the next generation is agentic. Ask your current vendors: ”Can we build custom security agents that automate detection and response workflows?” If the answer requires a roadmap conversation, you are behind the curve that NAB is already building on.

4. Palantir Just Renewed Its AI Partnership with Stellantis for Five More Years. Renewals Tell You What Press Releases Cannot.

Palantir announced the renewal and expansion of its partnership with Stellantis for an additional five years. This story will not make the front page of any tech publication. There is no eye-popping valuation, no ”largest ever” superlative, no founder with a colorful backstory. And that is exactly why it matters.

Renewals are the most honest signal in enterprise technology. A company can sign a splashy initial deal for brand association, press coverage, or board pressure. But a five-year renewal means the technology delivered enough value that the customer wants five more years of it. Stellantis is one of the world's largest automakers. They have no shortage of AI vendors pitching them. They chose to stay with Palantir and expand the scope.

The pattern is worth watching across the industry. As AI moves from pilot phase to operational deployment, the signal to track is no longer ”who signed a deal” but ”who renewed one.” The companies that earn renewals have proven they can integrate with enterprise workflows, survive changes in executive leadership, and deliver value that the finance team can quantify. The companies that only have first-year deals may still be running on enthusiasm rather than evidence.

Here's what works: Add a renewal tracking metric to your AI vendor evaluation process. For every AI partnership your organization has, ask: ”Would we renew this for five more years?” If the answer is uncertain, the implementation is not delivering enough value. If the answer is yes, you have found a real AI partner, not just a technology vendor.

5. A Nuclear Energy Startup Just Expanded a $200 Million Partnership to Power AI Data Centers. The Energy Problem Nobody Wants to Headline.

Oklo and Blykalla announced the expansion of their strategic partnership with up to $200 million in U.S. investment, positioning nuclear microreactors as the power source for next-generation AI data centers. Blykalla, a Swedish nuclear startup, builds compact reactors designed for industrial applications. Oklo is bringing that technology to the U.S. market with significant backing.

The energy problem in AI is no longer theoretical. Every hyperscaler's expansion roadmap is constrained by power availability. New data centers require dedicated power infrastructure that takes years to permit and build. Nuclear microreactors offer a path around that bottleneck: compact, always-on power generation that can be deployed at or near the data center site, without depending on the grid.

This is the infrastructure layer that rarely gets attention but determines whether AI scales or stalls. The models keep getting bigger. The training runs keep getting longer. The inference workloads keep multiplying. And every one of those trends increases power demand. The companies that secure dedicated, reliable energy sources will have a structural advantage over competitors who are fighting for grid capacity alongside every other industry.

Here's what works: Run an energy dependency audit on your AI infrastructure roadmap. Ask: ”If grid capacity in our primary data center markets becomes constrained, what is our fallback?” If the answer is ”move to another region,” that is a delay strategy, not a solution. The companies tracking nuclear, geothermal, and dedicated power solutions now will avoid the scramble later.

6. The Best AI in the World Just Scored 0.37 Percent on a New Benchmark. Read That Number Again.

A new benchmark has emerged where the most capable AI systems in the world score 0.37%. Not 37%. Not 3.7%. Zero point three seven percent. While billions are being deployed into AI across pharma, finance, security, and space, this benchmark is a cold reminder that there are entire categories of reasoning where the most advanced models are functionally helpless.

The timing could not be more instructive. In the same week that Eli Lilly signed a $2.75 billion AI deal and Starcloud reached unicorn status building orbital data centers, a simple test showed that the models powering these investments still have fundamental limitations. This is not a contradiction. It is a calibration. AI excels in specific, well-defined domains: pattern recognition in drug compounds, optimization in manufacturing logistics, anomaly detection in security data. It fails when asked to reason beyond those trained domains.

For enterprise leaders, this is the most useful data point of the week. The companies that succeed with AI are the ones that deploy it where it already works, not where they hope it will work someday. Lilly is not asking AI to reason about philosophy. They are asking it to identify drug candidates, a task where AI has demonstrated capability. The benchmark score is a warning against applying AI to tasks that require the kind of reasoning it cannot yet perform.

Here's what works: Before greenlighting any AI initiative, match the task to AI's proven capabilities. Ask: ”Is this a pattern recognition problem, an optimization problem, or a reasoning problem?” If it is the first two, AI is your accelerator. If it is the third, you need human judgment with AI assistance, not AI autonomy. The 0.37% score is your calibration tool.

Signal vs. Noise

🟢 Signal: Information security infrastructure is gaining foundational importance faster than any other concept this week. Security policy, data security, and incident response all surged in structural influence across the articles we track, with security-related topics rising nearly 4,000% in importance. This is not hype. When banks are building 30TB-per-day agentic security platforms and compliance mentions are compounding (20 GDPR references, 16 CCPA references in a single day), the market is telling you that security infrastructure is becoming the prerequisite for everything else. Companies without modern security foundations will find their AI ambitions blocked by compliance requirements and threat exposure, not by technology limitations.

🟢 Signal: AI in drug discovery has crossed from emerging to mainstream in a single deal cycle. Eli Lilly's $2.75 billion commitment, combined with Insilico's 28 AI-developed drugs and 50% stock surge, signals that pharma-AI is no longer speculative. The trend data confirms it: AI in drug discovery appears simultaneously in emerging, growing, and mainstream lifecycle stages, meaning different segments of the industry are at different adoption points, but the overall direction is clear and accelerating.

🔴 Noise: AI governance discourse continues to attract attention while its real-world influence is flat or declining. AI governance appeared in multiple articles this week, but its structural influence actually declined. The gap between governance talk and governance action is widening. Organizations are discussing frameworks without implementing them. Until governance translates into enforceable policy, operational budgets, and measurable outcomes, it remains noise, not signal.

From the 190K

AI Just Deployed Into Six Industries in 48 Hours. No One Publication Covered More Than One.

We scanned 190,000 articles this week. Here is the pattern that only emerges at scale:

In the same 48-hour window, Eli Lilly signed a $2.75 billion AI drug discovery deal, Starcloud raised $170 million for orbital AI data centers, Oklo expanded a $200 million nuclear energy partnership for AI power, NAB started co-designing an agentic security platform, Palantir renewed with Stellantis for five more years, and FactSet partnered with Finster AI to automate banking workflows. Pharma, space, nuclear energy, banking security, automotive, financial services. Six verticals, six significant deals, 48 hours.

No single publication covered more than one of these stories. The pharma press covered Lilly. The fintech press covered FactSet. The energy press covered Oklo. The pattern, that AI has simultaneously crossed from experimental to operational across six unrelated industries in a single week, only becomes visible when you read across all of them. This is what verticalization looks like at scale: not one industry adopting AI, but all of them, independently arriving at the same conclusion.

🔍 Below the surface: A German consortium called ASCEND just received €30 million to use AI for catalyst discovery in clean energy. Zero tech headlines. Zero VC coverage. But when AI starts accelerating materials science for green hydrogen and chemical processes, the clean energy timeline changes. The hype machine has not figured out how to make catalyst discovery sexy yet, which usually means it actually works.

By The Numbers

• $2.75 billion: Eli Lilly's deal with Insilico Medicine for AI drug discovery. The largest pharma-AI partnership ever announced.
• 28 AI-developed drugs: Insilico Medicine's portfolio, with nearly half already in clinical-stage trials. This is not a research lab. It is a pipeline.
• $170 million: Starcloud's Series A for orbital AI data centers, reaching unicorn status at $1.1 billion. The infrastructure frontier just left the atmosphere.
• 30 terabytes per day: The volume of security data NAB ingests daily, driving the need for agentic security platforms that no human team can monitor manually.
• $200 million: Oklo and Blykalla's expanded nuclear energy partnership for AI data center power. The energy bottleneck is attracting serious capital.
• 0.37%: The best AI score on a new reasoning benchmark. A reality check wrapped in a decimal point.
• 20 GDPR mentions: In a single day's articles, with CCPA at 16 and HIPAA at 5. Regulatory density is not easing.
• 50% stock surge: Insilico Medicine's gain since its Hong Kong IPO in December. The market is pricing in AI drug discovery as proven, not speculative.

Deep Dive: The Verticalization of AI, and Why the Biggest Money Just Moved Industry-Specific

You know that moment when a DJ stops playing festival anthems and starts reading the specific room? The crowd shifts. The energy gets more precise. The connection gets deeper. That is exactly what is happening in AI right now. The era of ”AI for everything” is giving way to ”AI for this specific problem, in this specific industry, with this specific data.”

The Horizontal Hangover

For three years, the AI market ran on horizontal plays. Build a foundation model. Make it work for everyone. Sell it to every industry. The valuations reflected that ambition. But the 0.37% benchmark score tells you what the horizontal approach ran into: general-purpose AI is extraordinarily capable in some domains and functionally useless in others. You cannot build a $2.75 billion pharma deal on general-purpose reasoning. You build it on a platform that has already produced 28 drug candidates and put half of them into clinical trials. Specificity is the product.

The Vertical Payoff

The deals this week are not random. Eli Lilly chose Insilico because it understands drug discovery at the molecular level. Palantir retained Stellantis because five years of automotive data integration proved the value. NAB is co-designing a security platform because 30TB of daily threat data requires tools built for banking-scale security, not retrofitted enterprise software. Each of these companies picked an AI partner that speaks their industry's language, not a generic platform that promises to learn it.

The Infrastructure Fork

The infrastructure layer is forking along the same vertical lines. Starcloud is building orbital data centers because certain AI workloads need more power than terrestrial grids can supply. Oklo is deploying nuclear microreactors because always-on AI inference needs always-on power. FactSet partnered with Finster AI because banking workflow automation requires financial domain expertise, not just language model capability. The infrastructure is becoming industry-specific too, and that changes the investment calculus for everyone.

What Actually Works

1. Audit your AI by vertical depth. List every AI initiative and score it: is the AI platform built for your industry, or is it a general tool you are adapting? The adapted ones will hit capability ceilings first.
2. Prioritize vendors with industry portfolios. Insilico did not get a $2.75 billion deal because it had the best model. It got the deal because it had 28 drugs in its pipeline. Demand evidence of industry-specific output, not model benchmarks.
3. Watch renewal patterns. Five-year renewals like Palantir-Stellantis are the strongest validation signal in enterprise AI. Add ”renewal rate” to your vendor evaluation criteria.
4. Map your energy and infrastructure dependencies by vertical. The same AI workload has different infrastructure needs in pharma (compute-heavy discovery), finance (latency-sensitive inference), and security (volume-intensive ingestion). Plan accordingly.

You do not play the same set at a jazz club and a warehouse rave. The AI companies that will dominate the next decade are the ones who learned to read the specific room they are playing in. The horizontal era built the instruments. The vertical era is where the music actually gets made.

What's Coming

Pharma-AI Deals Will Accelerate Past the Proof-of-Concept Stage

Eli Lilly's $2.75 billion commitment to Insilico validates AI drug discovery at a scale that rewrites procurement conversations across the industry. With 28 AI-developed drugs and a 50% stock surge backing the thesis, expect at least two more billion-dollar pharma-AI partnerships in Q2 2026 as competitors rush to match Lilly's positioning. The window for first-mover advantage in pharma-AI is closing.

Agentic Security Platforms Will Force Legacy SIEM Replacement Timelines

NAB's design partnership for a 30TB-per-day agentic security platform signals the end of rule-based security monitoring at enterprise scale. When one of the world's largest banks decides that human analysts cannot keep pace with threat volume and co-designs the replacement, legacy vendors have 18 months to reinvent or become irrelevant. Watch for design partner announcements from other major banks in Q2.

The AI Energy Race Will Produce Its First Regulatory Collision

Oklo's $200 million nuclear expansion and Starcloud's orbital ambitions both sit outside existing data center regulation. Nuclear microreactors at data center sites and computing infrastructure in orbit will force regulatory frameworks to catch up. The companies that engage regulators early will write the rules. The ones that wait will comply with rules someone else designed.

For Your Team

Wednesday's meeting prompt: ”This week, Eli Lilly bet $2.75 billion that AI can find drugs faster than traditional research. A startup raised $170 million to put data centers in orbit. And the best AI in the world scored 0.37% on a new reasoning test. Here is the question: if AI is mature enough for pharma to bet billions on it but still scores below 1% on reasoning benchmarks, what does that tell us about where we should be deploying AI and where we should not?”

The Vertical AI Readiness Audit:

1. Map your industry-specific opportunities. Generic AI saves time. Vertical AI creates competitive advantages. List three problems unique to your industry that AI could address better than a horizontal tool.
2. Check your renewal signals. If your AI vendors are earning renewals, the technology works. If they are being replaced, it does not. Track which AI partnerships are entering their second year and beyond.
3. Budget for domain expertise, not just compute. Eli Lilly's deal works because Insilico understands drug discovery at the molecular level, not because they have bigger GPUs. Your AI investments need the same vertical depth.
4. Stress-test against the 0.37% reality. If your AI use case requires reasoning at a level where the best models score below 1%, you are building on unstable ground. Pick use cases that match current capabilities.

Share-worthy stat: Insilico Medicine has developed 28 drugs using AI, with nearly half already in clinical trials. Their stock is up 50% since their Hong Kong IPO in December. AI drug discovery is no longer a promise. It is a portfolio.

Go deeper: Track vertical AI deployment signals in real-time →

The Track of the Day

”Do not play what's there. Play what's not there.”
— Miles Davis

Today's set: ”So What” by Miles Davis. In 1959, Miles walked into a studio with a radical idea: strip jazz down to its essence. No complex chord changes. No showing off technique. Just modes, space, and listening. The result was Kind of Blue, the best-selling jazz album of all time. Every musician on that session was a virtuoso, but Miles told them to play less. The magic was in the restraint. AI is at that same crossroads right now. The technology can do a thousand things. The companies winning are the ones choosing to do one thing, for one industry, with surgical precision. Lilly picked drug discovery. NAB picked security. Palantir picked automotive. Your DJ signing off: find your one industry, your one problem, your one vertical. And play it like Miles played ”So What,” with everything you have and nothing you do not need.

Yves Mulkers, your data DJ, mixing 190,000 articles into the tracks that actually matter.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: March 31, 2026 | Curated by Yves Mulkers @ Ins7ghts